Linguistic typology is currently struggling with the problem of making statistically valid statements about non-independent observations of typological variation. With only six or seven thousand languages in the world, many of them undocumented and all heavily skewed towards a few huge families, generalizations based on stratified samples will only ever be possible on a coarse scale. We use bioinformatic methods which model evolutionary processes to make statistically sensitive, probabilistic measurements of typological variation within a phylogenetic tree, enabling us to make statistical statements about typological stability, and to measure subtle typological interdependencies.